Generating game ideas with large language models
Abstract
Large language models are increasingly used as tools for creative ideation in domains such as game design. While modern chat-based models are effective at producing clear and coherent ideas, concerns have been raised about their creative diversity due to alignment and safety training. Prompt engineering is often proposed to increase creativity, but the relative importance of model choice and prompting strategy remains unclear, especially when creativity is evaluated by humans. This thesis empirically investigates how different large language models and prompting strategies influence the creativity of AI-generated game ideas. The study compares a base language model (davinci-002) and a chat-based model trained with reinforcement learning from human feedback (GPT-4o), using both regular fewshot prompting and combinatorial prompting. A total of 200 game ideas were generated and evaluated by game design students in terms of novelty and value. The results show that model choice has a substantially stronger impact on perceived creativity than prompting strategy. GPT-4o consistently produced game ideas that were rated higher in both novelty and value, despite exhibiting lower semantic diversity. In contrast, combinatorial prompting led to only minor differences in creativity scores across models. These findings suggest that coherence and usefulness play a more important role in human creativity judgments than diversity alone. Overall, this thesis provides empirical evidence that selecting an appropriate model is more critical than designing complex prompts for creative game ideation. The results contribute to a clearer understanding of AI-assisted creativity and offer practical guidance for the use of large language models in game design.
Keywords
Not available
Specialisations
Similar theses
- On the Safety Alignment of Large Language Models Mengxuan Hu · 2026
- Large Language Models for Symbolic Music Understanding Meng Yang · 2026
- Advancing personalized education with large language models Μήκος, Γεώργιος Σ. ; Μήκος, Γεώργιος Σ. · University of Thessaly · 2025
- Educational question generation and evaluation with large language models Chun Yi Louis Bo · 2024
- Cognitive Behaviours of Large Language Models Bryan Wilie · 2026
- Value estimation in Large Language Models Naga Lakshmi Prathyusha Guduru · 2026
